[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82439-en":3,"doc-seo-82439-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},82439,7971461741311,"Ophelia","https://ap-avatar.wpscdn.com/avatar/74000253aff267980c6?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779345379180704826",8,"Research & Report","Semantic Pareto-DQN: A Multi-Objective Reinforcement Learning Framework for Financial Anomaly Detection","Financial anomaly detection faces extreme class imbalance that causes single-objective models to collapse into the majority class, failing to balance fraud interdiction against customer friction. To address this without distortive resampling, the Semantic Pareto-DQN introduces a multi-objective reinforcement learning framework. Heterogeneous transaction features are synthesized into natural-language narratives via large language models to form robust, scale-invariant states. A vector reward decouples financial efficacy, operational friction, and semantic discovery, and continuous Pareto-front mapping mitigates missed-anomaly versus false-positive asymmetries.","arXiv :2607 .0964 1v 1 [ cs .LG] 10 Jul 2026  \nSemantic Pareto-DQN: A Multi-Objective Reinforcement Learning Framework for Financial Anomaly Detection  \nCláudio Lúcio do Val Lopes 1 [0000-0003-1655-2283] and Lucca Machado da Silva 1 [0009-0002-7191-6756]  \nA3Data-[http://www.a3data.com.br](http://www.a3data.com.br)[ ](http://www.a3data.com.br)Correspondence to: [claudio.lucio@a3data.com.br](claudio.lucio@a3data.com.br)  \nAbstract. Financial anomaly detection suffers from extreme class imbalance, causing traditional single-objective algorithms to exhibit “fraud collapse”, defaulting to the majority class and failing to balance anomaly interdiction with customer friction. To overcome this without distortive data resampling, we propose the Semantic Pareto-DQN, a multi-objective reinforcement learning framework. Our approach synthesizes heterogeneous transaction features into cohesive natural-language narratives, encoded by large language models, thereby producing a robust, scale-invariant state representation. The agent optimizes a vectorial reward that explicitly decouples financial efficacy, operational friction, and semantic discovery. By mapping the continuous Pareto frontier, the system dynamically navigates the asymmetric costs of missed anomalies versus false positives. Empirical evaluations across E-Commerce fraud and UCI Credit datasets show that semantic Pareto-DQN successfully shatters the zero-recall trap. It achieves superior minority-class recall compared to scalarized baselines, providing an alternative to trade bounded operational friction for financial anomaly discovery.  \nKeywords: Reinformment Learning · Multi-objective optimization · Evaluation process · Fraud Detection  \n1 Introduction  \nFinancial anomaly detection, encompassing e-commerce fraud and credit card defaults, is characterized by extreme class imbalance and adversarial adaptation [2,7] . Some machine learning approaches, including deep artificial neural networks, has been deployed to maximize predictive accuracy and safeguard data privacy in fraud detection [3] . However, as highlighted in recent reviews, these standard supervised algorithms predominantly optimize a single scalar metric, typically maximizing the log-likelihood. In e-commerce environments characterized by extreme class imbalance, this singular mathematical focus structurally incentivizes the model to collapse onto the majority-class manifold. Consequently, these static classifiers are fundamentally ill-equipped to dynamically balance the  \n2 Cláudio L V Lopes and Lucca M. da Silva  \nnon-linear trade-offs between maximizing anomaly detection efficacy and minimizing customer friction.  \nTo mitigate this extreme class imbalance, traditional literature frequently relies on data-level resampling techniques (e.g., SMOTE or ADASYN) [19] . However, as detailed in Section 2, these methods often introduce fictitious samples that distort real-world decision boundaries. Avoiding these distribution alterations, our methodology focuses purely on an algorithmic intervention.  \nOvercoming the static nature of standard supervised classifiers, recent literature has formulated fraud detection as a sequential decision-making process governed by Reinforcement Learning (RL) [18,4,15] . By treating transactions as state transitions within a Markov Decision Process (MDP), RL agents learn a dynamic policy πθ that maximizes cumulative expected returns rather than merely minimizing instantaneous classification error. However, the operational goal of an enterprise fraud detection system is fundamentally multi-dimensional: rather than minimizing all statistical errors, it requires dynamically managing the asymmetric trade-offs between maximizing the interdiction of rare fraud (True Positives) and bounding the inevitable operational friction (False Positives) associated with high-sensitivity detection. Current state-of-the-art architectures attempt to navigate these constraints via scalarization and asymmetric r","cbCaiqSFE85X1NwB","https://ap.wps.com/l/cbCaiqSFE85X1NwB","pdf",429384,1,15,"English","en",105,"# Introduction\n## Class imbalance and fraud collapse\n## Limits of scalar metrics and resampling\n## Reinforcement learning and multi-dimensional fraud goals\n## Multi-objective reinforcement learning and Pareto optimization","[{\"question\":\"Why do traditional single-objective algorithms struggle with financial anomaly detection?\",\"answer\":\"Extreme class imbalance incentivizes models to default to the majority class, leading to “fraud collapse” and poor balancing between anomaly interdiction and customer friction.\"},{\"question\":\"What distinguishes Semantic Pareto-DQN from scalarized reward approaches?\",\"answer\":\"Semantic Pareto-DQN uses a vectorial reward to explicitly decouple financial efficacy, operational friction, and semantic discovery, instead of forcing objectives into a single scalar.\"},{\"question\":\"How does the framework address the missed-anomaly vs false-positive trade-off?\",\"answer\":\"It maps and navigates the continuous Pareto frontier, dynamically adjusting actions to handle asymmetric costs of missed anomalies versus false positives.\"}]",1784180383,38,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"semantic-pareto-dqn-a-multi-objective-reinforcement-learning-framework-for-financial-anomaly-detection","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/semantic-pareto-dqn-a-multi-objective-reinforcement-learning-framework-for-financial-anomaly-detection/82439/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why do traditional single-objective algorithms struggle with financial anomaly detection?","Question",{"text":75,"@type":76},"Extreme class imbalance incentivizes models to default to the majority class, leading to “fraud collapse” and poor balancing between anomaly interdiction and customer friction.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What distinguishes Semantic Pareto-DQN from scalarized reward approaches?",{"text":80,"@type":76},"Semantic Pareto-DQN uses a vectorial reward to explicitly decouple financial efficacy, operational friction, and semantic discovery, instead of forcing objectives into a single scalar.",{"name":82,"@type":73,"acceptedAnswer":83},"How does the framework address the missed-anomaly vs false-positive trade-off?",{"text":84,"@type":76},"It maps and navigates the continuous Pareto frontier, dynamically adjusting actions to handle asymmetric costs of missed anomalies versus false positives.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]